Nvidia has made significant strides in the field of agentic AI by launching new services and models, including the Nemotron family, which is based on Meta’s Llama. This family consists of three models: Nano, Super, and Ultra, optimized for various computing resources. The Llama Nemotron focuses on language tasks, while the Cosmos Nemotron is designed for vision-related projects. The models range from 4B parameters in Nano to 253B in Ultra, catering to tasks like instruction following, chat, function calling, coding, and math.
Rev Lebaredian, Nvidia's VP of Omniverse and simulation technology, emphasized that these models serve as a digital workforce, enhancing agentic AI capabilities. The Nemotron models are accessible as hosted APIs on platforms like Hugging Face and Nvidia’s website, and enterprises can utilize them through Nvidia's AI Enterprise software.
The trend of AI agents gained momentum in 2024, with major companies like Salesforce, ServiceNow, AWS, and Microsoft recognizing their potential in enterprise workflows. Nvidia has also ventured into AI orchestration, providing blueprints that guide agents in specific tasks. Collaborations with orchestration companies such as LangChain and CrewAI have resulted in tailored blueprints for various applications, including code documentation and structured report generation.
Nvidia's new PDF-to-podcast blueprint aims to convert PDF information into audio, competing with Google’s NotebookLM. Additionally, a blueprint for summarizing videos is in development. The introduction of Nvidia Launchables allows developers to test and run blueprints easily, positioning orchestration as a key focus for enterprises in 2025 as they navigate multi-agent production.
